The problem

Stochastic generation buys you variety at the cost of repeatability. Run it twice and you get two different things, and the first one is gone for good. Embroidery wants the opposite discipline. A machine follows a stitch plan literally: needle down here, thread to there, no interpretation. So the design has to be exact, physically realisable, and recoverable on demand.

That is the tension worth solving. I wanted the surprise of a generative process and the determinism of a manufacturing file living in the same artefact: a design no one has seen before that I can nonetheless reproduce stitch for stitch.

What I built

Seedstitch is a generative engine in Python that reads an image and grows an embroidery design out of it. The randomness is seeded, so a seed plus an image always resolve to the same stitch plan, down to the last needle drop. Change the seed and you get a different design from the same picture. Keep the seed and you can recover any design you have ever generated.

  • Seeded, not merely random. Every stochastic decision draws from one seeded stream, so the output is a pure function of (image, seed). Reproducibility is built in, not bolted on.
  • One of one. A given seed yields a design that is unique to it. The engine generates an original each run rather than recolouring a fixed template.
  • Machine-stitchable by construction. The output is an ordered stitch path, the sequence a needle can actually follow, not a picture that still needs interpreting into thread.
  • Verified end to end. Reproducibility is checked the whole way through: same seed in, identical stitch plan out, confirmed rather than assumed.
open ⤢
An input image and a seed feed a stochastic generator that grows an ordered stitch path the machine can follow. The seed makes it deterministic: the same seed reproduces the same stitches, and each seed yields a one-of-one design.

Outcome

A working engine, shipped as an MVP. It takes a real image and a seed and produces a stitch plan a machine can run, with reproducibility verified end to end. The interesting result is the one that sounds contradictory at first: a generative process whose output is both genuinely one of a kind and perfectly repeatable.

Tech

Python. A seeded stochastic generator that maps an image onto an ordered, machine-stitchable stitch path, with end-to-end reproducibility checks tying a seed to its exact stitch plan.